xref: /openbmc/linux/arch/arm64/kernel/head.S (revision 20e2fc42)
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Low-level CPU initialisation
4 * Based on arch/arm/kernel/head.S
5 *
6 * Copyright (C) 1994-2002 Russell King
7 * Copyright (C) 2003-2012 ARM Ltd.
8 * Authors:	Catalin Marinas <catalin.marinas@arm.com>
9 *		Will Deacon <will.deacon@arm.com>
10 */
11
12#include <linux/linkage.h>
13#include <linux/init.h>
14#include <linux/irqchip/arm-gic-v3.h>
15
16#include <asm/assembler.h>
17#include <asm/boot.h>
18#include <asm/ptrace.h>
19#include <asm/asm-offsets.h>
20#include <asm/cache.h>
21#include <asm/cputype.h>
22#include <asm/elf.h>
23#include <asm/image.h>
24#include <asm/kernel-pgtable.h>
25#include <asm/kvm_arm.h>
26#include <asm/memory.h>
27#include <asm/pgtable-hwdef.h>
28#include <asm/pgtable.h>
29#include <asm/page.h>
30#include <asm/smp.h>
31#include <asm/sysreg.h>
32#include <asm/thread_info.h>
33#include <asm/virt.h>
34
35#include "efi-header.S"
36
37#define __PHYS_OFFSET	(KERNEL_START - TEXT_OFFSET)
38
39#if (TEXT_OFFSET & 0xfff) != 0
40#error TEXT_OFFSET must be at least 4KB aligned
41#elif (PAGE_OFFSET & 0x1fffff) != 0
42#error PAGE_OFFSET must be at least 2MB aligned
43#elif TEXT_OFFSET > 0x1fffff
44#error TEXT_OFFSET must be less than 2MB
45#endif
46
47/*
48 * Kernel startup entry point.
49 * ---------------------------
50 *
51 * The requirements are:
52 *   MMU = off, D-cache = off, I-cache = on or off,
53 *   x0 = physical address to the FDT blob.
54 *
55 * This code is mostly position independent so you call this at
56 * __pa(PAGE_OFFSET + TEXT_OFFSET).
57 *
58 * Note that the callee-saved registers are used for storing variables
59 * that are useful before the MMU is enabled. The allocations are described
60 * in the entry routines.
61 */
62	__HEAD
63_head:
64	/*
65	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
66	 */
67#ifdef CONFIG_EFI
68	/*
69	 * This add instruction has no meaningful effect except that
70	 * its opcode forms the magic "MZ" signature required by UEFI.
71	 */
72	add	x13, x18, #0x16
73	b	stext
74#else
75	b	stext				// branch to kernel start, magic
76	.long	0				// reserved
77#endif
78	le64sym	_kernel_offset_le		// Image load offset from start of RAM, little-endian
79	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
80	le64sym	_kernel_flags_le		// Informative flags, little-endian
81	.quad	0				// reserved
82	.quad	0				// reserved
83	.quad	0				// reserved
84	.ascii	ARM64_IMAGE_MAGIC		// Magic number
85#ifdef CONFIG_EFI
86	.long	pe_header - _head		// Offset to the PE header.
87
88pe_header:
89	__EFI_PE_HEADER
90#else
91	.long	0				// reserved
92#endif
93
94	__INIT
95
96	/*
97	 * The following callee saved general purpose registers are used on the
98	 * primary lowlevel boot path:
99	 *
100	 *  Register   Scope                      Purpose
101	 *  x21        stext() .. start_kernel()  FDT pointer passed at boot in x0
102	 *  x23        stext() .. start_kernel()  physical misalignment/KASLR offset
103	 *  x28        __create_page_tables()     callee preserved temp register
104	 *  x19/x20    __primary_switch()         callee preserved temp registers
105	 *  x24        __primary_switch() .. relocate_kernel()
106	 *                                        current RELR displacement
107	 */
108ENTRY(stext)
109	bl	preserve_boot_args
110	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
111	adrp	x23, __PHYS_OFFSET
112	and	x23, x23, MIN_KIMG_ALIGN - 1	// KASLR offset, defaults to 0
113	bl	set_cpu_boot_mode_flag
114	bl	__create_page_tables
115	/*
116	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
117	 * details.
118	 * On return, the CPU will be ready for the MMU to be turned on and
119	 * the TCR will have been set.
120	 */
121	bl	__cpu_setup			// initialise processor
122	b	__primary_switch
123ENDPROC(stext)
124
125/*
126 * Preserve the arguments passed by the bootloader in x0 .. x3
127 */
128preserve_boot_args:
129	mov	x21, x0				// x21=FDT
130
131	adr_l	x0, boot_args			// record the contents of
132	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
133	stp	x2, x3, [x0, #16]
134
135	dmb	sy				// needed before dc ivac with
136						// MMU off
137
138	mov	x1, #0x20			// 4 x 8 bytes
139	b	__inval_dcache_area		// tail call
140ENDPROC(preserve_boot_args)
141
142/*
143 * Macro to create a table entry to the next page.
144 *
145 *	tbl:	page table address
146 *	virt:	virtual address
147 *	shift:	#imm page table shift
148 *	ptrs:	#imm pointers per table page
149 *
150 * Preserves:	virt
151 * Corrupts:	ptrs, tmp1, tmp2
152 * Returns:	tbl -> next level table page address
153 */
154	.macro	create_table_entry, tbl, virt, shift, ptrs, tmp1, tmp2
155	add	\tmp1, \tbl, #PAGE_SIZE
156	phys_to_pte \tmp2, \tmp1
157	orr	\tmp2, \tmp2, #PMD_TYPE_TABLE	// address of next table and entry type
158	lsr	\tmp1, \virt, #\shift
159	sub	\ptrs, \ptrs, #1
160	and	\tmp1, \tmp1, \ptrs		// table index
161	str	\tmp2, [\tbl, \tmp1, lsl #3]
162	add	\tbl, \tbl, #PAGE_SIZE		// next level table page
163	.endm
164
165/*
166 * Macro to populate page table entries, these entries can be pointers to the next level
167 * or last level entries pointing to physical memory.
168 *
169 *	tbl:	page table address
170 *	rtbl:	pointer to page table or physical memory
171 *	index:	start index to write
172 *	eindex:	end index to write - [index, eindex] written to
173 *	flags:	flags for pagetable entry to or in
174 *	inc:	increment to rtbl between each entry
175 *	tmp1:	temporary variable
176 *
177 * Preserves:	tbl, eindex, flags, inc
178 * Corrupts:	index, tmp1
179 * Returns:	rtbl
180 */
181	.macro populate_entries, tbl, rtbl, index, eindex, flags, inc, tmp1
182.Lpe\@:	phys_to_pte \tmp1, \rtbl
183	orr	\tmp1, \tmp1, \flags	// tmp1 = table entry
184	str	\tmp1, [\tbl, \index, lsl #3]
185	add	\rtbl, \rtbl, \inc	// rtbl = pa next level
186	add	\index, \index, #1
187	cmp	\index, \eindex
188	b.ls	.Lpe\@
189	.endm
190
191/*
192 * Compute indices of table entries from virtual address range. If multiple entries
193 * were needed in the previous page table level then the next page table level is assumed
194 * to be composed of multiple pages. (This effectively scales the end index).
195 *
196 *	vstart:	virtual address of start of range
197 *	vend:	virtual address of end of range
198 *	shift:	shift used to transform virtual address into index
199 *	ptrs:	number of entries in page table
200 *	istart:	index in table corresponding to vstart
201 *	iend:	index in table corresponding to vend
202 *	count:	On entry: how many extra entries were required in previous level, scales
203 *			  our end index.
204 *		On exit: returns how many extra entries required for next page table level
205 *
206 * Preserves:	vstart, vend, shift, ptrs
207 * Returns:	istart, iend, count
208 */
209	.macro compute_indices, vstart, vend, shift, ptrs, istart, iend, count
210	lsr	\iend, \vend, \shift
211	mov	\istart, \ptrs
212	sub	\istart, \istart, #1
213	and	\iend, \iend, \istart	// iend = (vend >> shift) & (ptrs - 1)
214	mov	\istart, \ptrs
215	mul	\istart, \istart, \count
216	add	\iend, \iend, \istart	// iend += (count - 1) * ptrs
217					// our entries span multiple tables
218
219	lsr	\istart, \vstart, \shift
220	mov	\count, \ptrs
221	sub	\count, \count, #1
222	and	\istart, \istart, \count
223
224	sub	\count, \iend, \istart
225	.endm
226
227/*
228 * Map memory for specified virtual address range. Each level of page table needed supports
229 * multiple entries. If a level requires n entries the next page table level is assumed to be
230 * formed from n pages.
231 *
232 *	tbl:	location of page table
233 *	rtbl:	address to be used for first level page table entry (typically tbl + PAGE_SIZE)
234 *	vstart:	start address to map
235 *	vend:	end address to map - we map [vstart, vend]
236 *	flags:	flags to use to map last level entries
237 *	phys:	physical address corresponding to vstart - physical memory is contiguous
238 *	pgds:	the number of pgd entries
239 *
240 * Temporaries:	istart, iend, tmp, count, sv - these need to be different registers
241 * Preserves:	vstart, vend, flags
242 * Corrupts:	tbl, rtbl, istart, iend, tmp, count, sv
243 */
244	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, pgds, istart, iend, tmp, count, sv
245	add \rtbl, \tbl, #PAGE_SIZE
246	mov \sv, \rtbl
247	mov \count, #0
248	compute_indices \vstart, \vend, #PGDIR_SHIFT, \pgds, \istart, \iend, \count
249	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
250	mov \tbl, \sv
251	mov \sv, \rtbl
252
253#if SWAPPER_PGTABLE_LEVELS > 3
254	compute_indices \vstart, \vend, #PUD_SHIFT, #PTRS_PER_PUD, \istart, \iend, \count
255	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
256	mov \tbl, \sv
257	mov \sv, \rtbl
258#endif
259
260#if SWAPPER_PGTABLE_LEVELS > 2
261	compute_indices \vstart, \vend, #SWAPPER_TABLE_SHIFT, #PTRS_PER_PMD, \istart, \iend, \count
262	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
263	mov \tbl, \sv
264#endif
265
266	compute_indices \vstart, \vend, #SWAPPER_BLOCK_SHIFT, #PTRS_PER_PTE, \istart, \iend, \count
267	bic \count, \phys, #SWAPPER_BLOCK_SIZE - 1
268	populate_entries \tbl, \count, \istart, \iend, \flags, #SWAPPER_BLOCK_SIZE, \tmp
269	.endm
270
271/*
272 * Setup the initial page tables. We only setup the barest amount which is
273 * required to get the kernel running. The following sections are required:
274 *   - identity mapping to enable the MMU (low address, TTBR0)
275 *   - first few MB of the kernel linear mapping to jump to once the MMU has
276 *     been enabled
277 */
278__create_page_tables:
279	mov	x28, lr
280
281	/*
282	 * Invalidate the init page tables to avoid potential dirty cache lines
283	 * being evicted. Other page tables are allocated in rodata as part of
284	 * the kernel image, and thus are clean to the PoC per the boot
285	 * protocol.
286	 */
287	adrp	x0, init_pg_dir
288	adrp	x1, init_pg_end
289	sub	x1, x1, x0
290	bl	__inval_dcache_area
291
292	/*
293	 * Clear the init page tables.
294	 */
295	adrp	x0, init_pg_dir
296	adrp	x1, init_pg_end
297	sub	x1, x1, x0
2981:	stp	xzr, xzr, [x0], #16
299	stp	xzr, xzr, [x0], #16
300	stp	xzr, xzr, [x0], #16
301	stp	xzr, xzr, [x0], #16
302	subs	x1, x1, #64
303	b.ne	1b
304
305	mov	x7, SWAPPER_MM_MMUFLAGS
306
307	/*
308	 * Create the identity mapping.
309	 */
310	adrp	x0, idmap_pg_dir
311	adrp	x3, __idmap_text_start		// __pa(__idmap_text_start)
312
313#ifdef CONFIG_ARM64_VA_BITS_52
314	mrs_s	x6, SYS_ID_AA64MMFR2_EL1
315	and	x6, x6, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
316	mov	x5, #52
317	cbnz	x6, 1f
318#endif
319	mov	x5, #VA_BITS_MIN
3201:
321	adr_l	x6, vabits_actual
322	str	x5, [x6]
323	dmb	sy
324	dc	ivac, x6		// Invalidate potentially stale cache line
325
326	/*
327	 * VA_BITS may be too small to allow for an ID mapping to be created
328	 * that covers system RAM if that is located sufficiently high in the
329	 * physical address space. So for the ID map, use an extended virtual
330	 * range in that case, and configure an additional translation level
331	 * if needed.
332	 *
333	 * Calculate the maximum allowed value for TCR_EL1.T0SZ so that the
334	 * entire ID map region can be mapped. As T0SZ == (64 - #bits used),
335	 * this number conveniently equals the number of leading zeroes in
336	 * the physical address of __idmap_text_end.
337	 */
338	adrp	x5, __idmap_text_end
339	clz	x5, x5
340	cmp	x5, TCR_T0SZ(VA_BITS)	// default T0SZ small enough?
341	b.ge	1f			// .. then skip VA range extension
342
343	adr_l	x6, idmap_t0sz
344	str	x5, [x6]
345	dmb	sy
346	dc	ivac, x6		// Invalidate potentially stale cache line
347
348#if (VA_BITS < 48)
349#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
350#define EXTRA_PTRS	(1 << (PHYS_MASK_SHIFT - EXTRA_SHIFT))
351
352	/*
353	 * If VA_BITS < 48, we have to configure an additional table level.
354	 * First, we have to verify our assumption that the current value of
355	 * VA_BITS was chosen such that all translation levels are fully
356	 * utilised, and that lowering T0SZ will always result in an additional
357	 * translation level to be configured.
358	 */
359#if VA_BITS != EXTRA_SHIFT
360#error "Mismatch between VA_BITS and page size/number of translation levels"
361#endif
362
363	mov	x4, EXTRA_PTRS
364	create_table_entry x0, x3, EXTRA_SHIFT, x4, x5, x6
365#else
366	/*
367	 * If VA_BITS == 48, we don't have to configure an additional
368	 * translation level, but the top-level table has more entries.
369	 */
370	mov	x4, #1 << (PHYS_MASK_SHIFT - PGDIR_SHIFT)
371	str_l	x4, idmap_ptrs_per_pgd, x5
372#endif
3731:
374	ldr_l	x4, idmap_ptrs_per_pgd
375	mov	x5, x3				// __pa(__idmap_text_start)
376	adr_l	x6, __idmap_text_end		// __pa(__idmap_text_end)
377
378	map_memory x0, x1, x3, x6, x7, x3, x4, x10, x11, x12, x13, x14
379
380	/*
381	 * Map the kernel image (starting with PHYS_OFFSET).
382	 */
383	adrp	x0, init_pg_dir
384	mov_q	x5, KIMAGE_VADDR + TEXT_OFFSET	// compile time __va(_text)
385	add	x5, x5, x23			// add KASLR displacement
386	mov	x4, PTRS_PER_PGD
387	adrp	x6, _end			// runtime __pa(_end)
388	adrp	x3, _text			// runtime __pa(_text)
389	sub	x6, x6, x3			// _end - _text
390	add	x6, x6, x5			// runtime __va(_end)
391
392	map_memory x0, x1, x5, x6, x7, x3, x4, x10, x11, x12, x13, x14
393
394	/*
395	 * Since the page tables have been populated with non-cacheable
396	 * accesses (MMU disabled), invalidate the idmap and swapper page
397	 * tables again to remove any speculatively loaded cache lines.
398	 */
399	adrp	x0, idmap_pg_dir
400	adrp	x1, init_pg_end
401	sub	x1, x1, x0
402	dmb	sy
403	bl	__inval_dcache_area
404
405	ret	x28
406ENDPROC(__create_page_tables)
407	.ltorg
408
409/*
410 * The following fragment of code is executed with the MMU enabled.
411 *
412 *   x0 = __PHYS_OFFSET
413 */
414__primary_switched:
415	adrp	x4, init_thread_union
416	add	sp, x4, #THREAD_SIZE
417	adr_l	x5, init_task
418	msr	sp_el0, x5			// Save thread_info
419
420	adr_l	x8, vectors			// load VBAR_EL1 with virtual
421	msr	vbar_el1, x8			// vector table address
422	isb
423
424	stp	xzr, x30, [sp, #-16]!
425	mov	x29, sp
426
427	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
428
429	ldr_l	x4, kimage_vaddr		// Save the offset between
430	sub	x4, x4, x0			// the kernel virtual and
431	str_l	x4, kimage_voffset, x5		// physical mappings
432
433	// Clear BSS
434	adr_l	x0, __bss_start
435	mov	x1, xzr
436	adr_l	x2, __bss_stop
437	sub	x2, x2, x0
438	bl	__pi_memset
439	dsb	ishst				// Make zero page visible to PTW
440
441#ifdef CONFIG_KASAN
442	bl	kasan_early_init
443#endif
444#ifdef CONFIG_RANDOMIZE_BASE
445	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
446	b.ne	0f
447	mov	x0, x21				// pass FDT address in x0
448	bl	kaslr_early_init		// parse FDT for KASLR options
449	cbz	x0, 0f				// KASLR disabled? just proceed
450	orr	x23, x23, x0			// record KASLR offset
451	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
452	ret					// to __primary_switch()
4530:
454#endif
455	add	sp, sp, #16
456	mov	x29, #0
457	mov	x30, #0
458	b	start_kernel
459ENDPROC(__primary_switched)
460
461/*
462 * end early head section, begin head code that is also used for
463 * hotplug and needs to have the same protections as the text region
464 */
465	.section ".idmap.text","awx"
466
467ENTRY(kimage_vaddr)
468	.quad		_text - TEXT_OFFSET
469EXPORT_SYMBOL(kimage_vaddr)
470
471/*
472 * If we're fortunate enough to boot at EL2, ensure that the world is
473 * sane before dropping to EL1.
474 *
475 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
476 * booted in EL1 or EL2 respectively.
477 */
478ENTRY(el2_setup)
479	msr	SPsel, #1			// We want to use SP_EL{1,2}
480	mrs	x0, CurrentEL
481	cmp	x0, #CurrentEL_EL2
482	b.eq	1f
483	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
484	msr	sctlr_el1, x0
485	mov	w0, #BOOT_CPU_MODE_EL1		// This cpu booted in EL1
486	isb
487	ret
488
4891:	mov_q	x0, (SCTLR_EL2_RES1 | ENDIAN_SET_EL2)
490	msr	sctlr_el2, x0
491
492#ifdef CONFIG_ARM64_VHE
493	/*
494	 * Check for VHE being present. For the rest of the EL2 setup,
495	 * x2 being non-zero indicates that we do have VHE, and that the
496	 * kernel is intended to run at EL2.
497	 */
498	mrs	x2, id_aa64mmfr1_el1
499	ubfx	x2, x2, #ID_AA64MMFR1_VHE_SHIFT, #4
500#else
501	mov	x2, xzr
502#endif
503
504	/* Hyp configuration. */
505	mov_q	x0, HCR_HOST_NVHE_FLAGS
506	cbz	x2, set_hcr
507	mov_q	x0, HCR_HOST_VHE_FLAGS
508set_hcr:
509	msr	hcr_el2, x0
510	isb
511
512	/*
513	 * Allow Non-secure EL1 and EL0 to access physical timer and counter.
514	 * This is not necessary for VHE, since the host kernel runs in EL2,
515	 * and EL0 accesses are configured in the later stage of boot process.
516	 * Note that when HCR_EL2.E2H == 1, CNTHCTL_EL2 has the same bit layout
517	 * as CNTKCTL_EL1, and CNTKCTL_EL1 accessing instructions are redefined
518	 * to access CNTHCTL_EL2. This allows the kernel designed to run at EL1
519	 * to transparently mess with the EL0 bits via CNTKCTL_EL1 access in
520	 * EL2.
521	 */
522	cbnz	x2, 1f
523	mrs	x0, cnthctl_el2
524	orr	x0, x0, #3			// Enable EL1 physical timers
525	msr	cnthctl_el2, x0
5261:
527	msr	cntvoff_el2, xzr		// Clear virtual offset
528
529#ifdef CONFIG_ARM_GIC_V3
530	/* GICv3 system register access */
531	mrs	x0, id_aa64pfr0_el1
532	ubfx	x0, x0, #ID_AA64PFR0_GIC_SHIFT, #4
533	cbz	x0, 3f
534
535	mrs_s	x0, SYS_ICC_SRE_EL2
536	orr	x0, x0, #ICC_SRE_EL2_SRE	// Set ICC_SRE_EL2.SRE==1
537	orr	x0, x0, #ICC_SRE_EL2_ENABLE	// Set ICC_SRE_EL2.Enable==1
538	msr_s	SYS_ICC_SRE_EL2, x0
539	isb					// Make sure SRE is now set
540	mrs_s	x0, SYS_ICC_SRE_EL2		// Read SRE back,
541	tbz	x0, #0, 3f			// and check that it sticks
542	msr_s	SYS_ICH_HCR_EL2, xzr		// Reset ICC_HCR_EL2 to defaults
543
5443:
545#endif
546
547	/* Populate ID registers. */
548	mrs	x0, midr_el1
549	mrs	x1, mpidr_el1
550	msr	vpidr_el2, x0
551	msr	vmpidr_el2, x1
552
553#ifdef CONFIG_COMPAT
554	msr	hstr_el2, xzr			// Disable CP15 traps to EL2
555#endif
556
557	/* EL2 debug */
558	mrs	x1, id_aa64dfr0_el1
559	sbfx	x0, x1, #ID_AA64DFR0_PMUVER_SHIFT, #4
560	cmp	x0, #1
561	b.lt	4f				// Skip if no PMU present
562	mrs	x0, pmcr_el0			// Disable debug access traps
563	ubfx	x0, x0, #11, #5			// to EL2 and allow access to
5644:
565	csel	x3, xzr, x0, lt			// all PMU counters from EL1
566
567	/* Statistical profiling */
568	ubfx	x0, x1, #ID_AA64DFR0_PMSVER_SHIFT, #4
569	cbz	x0, 7f				// Skip if SPE not present
570	cbnz	x2, 6f				// VHE?
571	mrs_s	x4, SYS_PMBIDR_EL1		// If SPE available at EL2,
572	and	x4, x4, #(1 << SYS_PMBIDR_EL1_P_SHIFT)
573	cbnz	x4, 5f				// then permit sampling of physical
574	mov	x4, #(1 << SYS_PMSCR_EL2_PCT_SHIFT | \
575		      1 << SYS_PMSCR_EL2_PA_SHIFT)
576	msr_s	SYS_PMSCR_EL2, x4		// addresses and physical counter
5775:
578	mov	x1, #(MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT)
579	orr	x3, x3, x1			// If we don't have VHE, then
580	b	7f				// use EL1&0 translation.
5816:						// For VHE, use EL2 translation
582	orr	x3, x3, #MDCR_EL2_TPMS		// and disable access from EL1
5837:
584	msr	mdcr_el2, x3			// Configure debug traps
585
586	/* LORegions */
587	mrs	x1, id_aa64mmfr1_el1
588	ubfx	x0, x1, #ID_AA64MMFR1_LOR_SHIFT, 4
589	cbz	x0, 1f
590	msr_s	SYS_LORC_EL1, xzr
5911:
592
593	/* Stage-2 translation */
594	msr	vttbr_el2, xzr
595
596	cbz	x2, install_el2_stub
597
598	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
599	isb
600	ret
601
602install_el2_stub:
603	/*
604	 * When VHE is not in use, early init of EL2 and EL1 needs to be
605	 * done here.
606	 * When VHE _is_ in use, EL1 will not be used in the host and
607	 * requires no configuration, and all non-hyp-specific EL2 setup
608	 * will be done via the _EL1 system register aliases in __cpu_setup.
609	 */
610	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
611	msr	sctlr_el1, x0
612
613	/* Coprocessor traps. */
614	mov	x0, #0x33ff
615	msr	cptr_el2, x0			// Disable copro. traps to EL2
616
617	/* SVE register access */
618	mrs	x1, id_aa64pfr0_el1
619	ubfx	x1, x1, #ID_AA64PFR0_SVE_SHIFT, #4
620	cbz	x1, 7f
621
622	bic	x0, x0, #CPTR_EL2_TZ		// Also disable SVE traps
623	msr	cptr_el2, x0			// Disable copro. traps to EL2
624	isb
625	mov	x1, #ZCR_ELx_LEN_MASK		// SVE: Enable full vector
626	msr_s	SYS_ZCR_EL2, x1			// length for EL1.
627
628	/* Hypervisor stub */
6297:	adr_l	x0, __hyp_stub_vectors
630	msr	vbar_el2, x0
631
632	/* spsr */
633	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
634		      PSR_MODE_EL1h)
635	msr	spsr_el2, x0
636	msr	elr_el2, lr
637	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
638	eret
639ENDPROC(el2_setup)
640
641/*
642 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
643 * in w0. See arch/arm64/include/asm/virt.h for more info.
644 */
645set_cpu_boot_mode_flag:
646	adr_l	x1, __boot_cpu_mode
647	cmp	w0, #BOOT_CPU_MODE_EL2
648	b.ne	1f
649	add	x1, x1, #4
6501:	str	w0, [x1]			// This CPU has booted in EL1
651	dmb	sy
652	dc	ivac, x1			// Invalidate potentially stale cache line
653	ret
654ENDPROC(set_cpu_boot_mode_flag)
655
656/*
657 * These values are written with the MMU off, but read with the MMU on.
658 * Writers will invalidate the corresponding address, discarding up to a
659 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
660 * sufficient alignment that the CWG doesn't overlap another section.
661 */
662	.pushsection ".mmuoff.data.write", "aw"
663/*
664 * We need to find out the CPU boot mode long after boot, so we need to
665 * store it in a writable variable.
666 *
667 * This is not in .bss, because we set it sufficiently early that the boot-time
668 * zeroing of .bss would clobber it.
669 */
670ENTRY(__boot_cpu_mode)
671	.long	BOOT_CPU_MODE_EL2
672	.long	BOOT_CPU_MODE_EL1
673/*
674 * The booting CPU updates the failed status @__early_cpu_boot_status,
675 * with MMU turned off.
676 */
677ENTRY(__early_cpu_boot_status)
678	.quad 	0
679
680	.popsection
681
682	/*
683	 * This provides a "holding pen" for platforms to hold all secondary
684	 * cores are held until we're ready for them to initialise.
685	 */
686ENTRY(secondary_holding_pen)
687	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
688	bl	set_cpu_boot_mode_flag
689	mrs	x0, mpidr_el1
690	mov_q	x1, MPIDR_HWID_BITMASK
691	and	x0, x0, x1
692	adr_l	x3, secondary_holding_pen_release
693pen:	ldr	x4, [x3]
694	cmp	x4, x0
695	b.eq	secondary_startup
696	wfe
697	b	pen
698ENDPROC(secondary_holding_pen)
699
700	/*
701	 * Secondary entry point that jumps straight into the kernel. Only to
702	 * be used where CPUs are brought online dynamically by the kernel.
703	 */
704ENTRY(secondary_entry)
705	bl	el2_setup			// Drop to EL1
706	bl	set_cpu_boot_mode_flag
707	b	secondary_startup
708ENDPROC(secondary_entry)
709
710secondary_startup:
711	/*
712	 * Common entry point for secondary CPUs.
713	 */
714	bl	__cpu_secondary_check52bitva
715	bl	__cpu_setup			// initialise processor
716	adrp	x1, swapper_pg_dir
717	bl	__enable_mmu
718	ldr	x8, =__secondary_switched
719	br	x8
720ENDPROC(secondary_startup)
721
722__secondary_switched:
723	adr_l	x5, vectors
724	msr	vbar_el1, x5
725	isb
726
727	adr_l	x0, secondary_data
728	ldr	x1, [x0, #CPU_BOOT_STACK]	// get secondary_data.stack
729	cbz	x1, __secondary_too_slow
730	mov	sp, x1
731	ldr	x2, [x0, #CPU_BOOT_TASK]
732	cbz	x2, __secondary_too_slow
733	msr	sp_el0, x2
734	mov	x29, #0
735	mov	x30, #0
736	b	secondary_start_kernel
737ENDPROC(__secondary_switched)
738
739__secondary_too_slow:
740	wfe
741	wfi
742	b	__secondary_too_slow
743ENDPROC(__secondary_too_slow)
744
745/*
746 * The booting CPU updates the failed status @__early_cpu_boot_status,
747 * with MMU turned off.
748 *
749 * update_early_cpu_boot_status tmp, status
750 *  - Corrupts tmp1, tmp2
751 *  - Writes 'status' to __early_cpu_boot_status and makes sure
752 *    it is committed to memory.
753 */
754
755	.macro	update_early_cpu_boot_status status, tmp1, tmp2
756	mov	\tmp2, #\status
757	adr_l	\tmp1, __early_cpu_boot_status
758	str	\tmp2, [\tmp1]
759	dmb	sy
760	dc	ivac, \tmp1			// Invalidate potentially stale cache line
761	.endm
762
763/*
764 * Enable the MMU.
765 *
766 *  x0  = SCTLR_EL1 value for turning on the MMU.
767 *  x1  = TTBR1_EL1 value
768 *
769 * Returns to the caller via x30/lr. This requires the caller to be covered
770 * by the .idmap.text section.
771 *
772 * Checks if the selected granule size is supported by the CPU.
773 * If it isn't, park the CPU
774 */
775ENTRY(__enable_mmu)
776	mrs	x2, ID_AA64MMFR0_EL1
777	ubfx	x2, x2, #ID_AA64MMFR0_TGRAN_SHIFT, 4
778	cmp	x2, #ID_AA64MMFR0_TGRAN_SUPPORTED
779	b.ne	__no_granule_support
780	update_early_cpu_boot_status 0, x2, x3
781	adrp	x2, idmap_pg_dir
782	phys_to_ttbr x1, x1
783	phys_to_ttbr x2, x2
784	msr	ttbr0_el1, x2			// load TTBR0
785	offset_ttbr1 x1, x3
786	msr	ttbr1_el1, x1			// load TTBR1
787	isb
788	msr	sctlr_el1, x0
789	isb
790	/*
791	 * Invalidate the local I-cache so that any instructions fetched
792	 * speculatively from the PoC are discarded, since they may have
793	 * been dynamically patched at the PoU.
794	 */
795	ic	iallu
796	dsb	nsh
797	isb
798	ret
799ENDPROC(__enable_mmu)
800
801ENTRY(__cpu_secondary_check52bitva)
802#ifdef CONFIG_ARM64_VA_BITS_52
803	ldr_l	x0, vabits_actual
804	cmp	x0, #52
805	b.ne	2f
806
807	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
808	and	x0, x0, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
809	cbnz	x0, 2f
810
811	update_early_cpu_boot_status \
812		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
8131:	wfe
814	wfi
815	b	1b
816
817#endif
8182:	ret
819ENDPROC(__cpu_secondary_check52bitva)
820
821__no_granule_support:
822	/* Indicate that this CPU can't boot and is stuck in the kernel */
823	update_early_cpu_boot_status \
824		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
8251:
826	wfe
827	wfi
828	b	1b
829ENDPROC(__no_granule_support)
830
831#ifdef CONFIG_RELOCATABLE
832__relocate_kernel:
833	/*
834	 * Iterate over each entry in the relocation table, and apply the
835	 * relocations in place.
836	 */
837	ldr	w9, =__rela_offset		// offset to reloc table
838	ldr	w10, =__rela_size		// size of reloc table
839
840	mov_q	x11, KIMAGE_VADDR		// default virtual offset
841	add	x11, x11, x23			// actual virtual offset
842	add	x9, x9, x11			// __va(.rela)
843	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
844
8450:	cmp	x9, x10
846	b.hs	1f
847	ldp	x12, x13, [x9], #24
848	ldr	x14, [x9, #-8]
849	cmp	w13, #R_AARCH64_RELATIVE
850	b.ne	0b
851	add	x14, x14, x23			// relocate
852	str	x14, [x12, x23]
853	b	0b
854
8551:
856#ifdef CONFIG_RELR
857	/*
858	 * Apply RELR relocations.
859	 *
860	 * RELR is a compressed format for storing relative relocations. The
861	 * encoded sequence of entries looks like:
862	 * [ AAAAAAAA BBBBBBB1 BBBBBBB1 ... AAAAAAAA BBBBBB1 ... ]
863	 *
864	 * i.e. start with an address, followed by any number of bitmaps. The
865	 * address entry encodes 1 relocation. The subsequent bitmap entries
866	 * encode up to 63 relocations each, at subsequent offsets following
867	 * the last address entry.
868	 *
869	 * The bitmap entries must have 1 in the least significant bit. The
870	 * assumption here is that an address cannot have 1 in lsb. Odd
871	 * addresses are not supported. Any odd addresses are stored in the RELA
872	 * section, which is handled above.
873	 *
874	 * Excluding the least significant bit in the bitmap, each non-zero
875	 * bit in the bitmap represents a relocation to be applied to
876	 * a corresponding machine word that follows the base address
877	 * word. The second least significant bit represents the machine
878	 * word immediately following the initial address, and each bit
879	 * that follows represents the next word, in linear order. As such,
880	 * a single bitmap can encode up to 63 relocations in a 64-bit object.
881	 *
882	 * In this implementation we store the address of the next RELR table
883	 * entry in x9, the address being relocated by the current address or
884	 * bitmap entry in x13 and the address being relocated by the current
885	 * bit in x14.
886	 *
887	 * Because addends are stored in place in the binary, RELR relocations
888	 * cannot be applied idempotently. We use x24 to keep track of the
889	 * currently applied displacement so that we can correctly relocate if
890	 * __relocate_kernel is called twice with non-zero displacements (i.e.
891	 * if there is both a physical misalignment and a KASLR displacement).
892	 */
893	ldr	w9, =__relr_offset		// offset to reloc table
894	ldr	w10, =__relr_size		// size of reloc table
895	add	x9, x9, x11			// __va(.relr)
896	add	x10, x9, x10			// __va(.relr) + sizeof(.relr)
897
898	sub	x15, x23, x24			// delta from previous offset
899	cbz	x15, 7f				// nothing to do if unchanged
900	mov	x24, x23			// save new offset
901
9022:	cmp	x9, x10
903	b.hs	7f
904	ldr	x11, [x9], #8
905	tbnz	x11, #0, 3f			// branch to handle bitmaps
906	add	x13, x11, x23
907	ldr	x12, [x13]			// relocate address entry
908	add	x12, x12, x15
909	str	x12, [x13], #8			// adjust to start of bitmap
910	b	2b
911
9123:	mov	x14, x13
9134:	lsr	x11, x11, #1
914	cbz	x11, 6f
915	tbz	x11, #0, 5f			// skip bit if not set
916	ldr	x12, [x14]			// relocate bit
917	add	x12, x12, x15
918	str	x12, [x14]
919
9205:	add	x14, x14, #8			// move to next bit's address
921	b	4b
922
9236:	/*
924	 * Move to the next bitmap's address. 8 is the word size, and 63 is the
925	 * number of significant bits in a bitmap entry.
926	 */
927	add	x13, x13, #(8 * 63)
928	b	2b
929
9307:
931#endif
932	ret
933
934ENDPROC(__relocate_kernel)
935#endif
936
937__primary_switch:
938#ifdef CONFIG_RANDOMIZE_BASE
939	mov	x19, x0				// preserve new SCTLR_EL1 value
940	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
941#endif
942
943	adrp	x1, init_pg_dir
944	bl	__enable_mmu
945#ifdef CONFIG_RELOCATABLE
946#ifdef CONFIG_RELR
947	mov	x24, #0				// no RELR displacement yet
948#endif
949	bl	__relocate_kernel
950#ifdef CONFIG_RANDOMIZE_BASE
951	ldr	x8, =__primary_switched
952	adrp	x0, __PHYS_OFFSET
953	blr	x8
954
955	/*
956	 * If we return here, we have a KASLR displacement in x23 which we need
957	 * to take into account by discarding the current kernel mapping and
958	 * creating a new one.
959	 */
960	pre_disable_mmu_workaround
961	msr	sctlr_el1, x20			// disable the MMU
962	isb
963	bl	__create_page_tables		// recreate kernel mapping
964
965	tlbi	vmalle1				// Remove any stale TLB entries
966	dsb	nsh
967
968	msr	sctlr_el1, x19			// re-enable the MMU
969	isb
970	ic	iallu				// flush instructions fetched
971	dsb	nsh				// via old mapping
972	isb
973
974	bl	__relocate_kernel
975#endif
976#endif
977	ldr	x8, =__primary_switched
978	adrp	x0, __PHYS_OFFSET
979	br	x8
980ENDPROC(__primary_switch)
981